Internship proposal
Evaluation of Techniques for Automatic Video Sequences Interpretation
performance evaluation, video analysis
| Internship Context and Goals |
The scope of the internship is in agreement with the ongoing research of the ORION group. It concerns automatic interpretation of video sequences pour human behavior recognition.
The ORION team has developed video sequence interpretation platform capable of detecting and tracking moving objects and recognizing certain behaviors that are predefined by experts in a database of scenarii. Especially, the platform has been tested for various applications such as metro and bank surveillance.
In addition to the a priori knowledge about the scenarii to be recognized and the environment observed, at each instant, the recognition process takes as input: (1) a set of individuals being tracked by a vision module and (2) a set of scenarii already recognized. The output of the process is a set of scenarii recognized for the current instant. A scenario is a model of the behavior of one or several individuals evolving in a scene in a given duration. A scenario is composed of a sequence of sub-scenarii meeting some temporal constraints.
The perceived real world is dynamic and contains uncertainty. Therefore, the image interpretation process is often subject to errors and difficult to evaluate. The Techno-Vision ETISEO project (Jan 2005 - 2007) aims at the evaluation of video interpretation techniques for video surveillance applications. The task of the ORION group consists of proposition of evaluation metrics, a tool for automatic comparison of video analysis results with ground truth, as well as execution of their own algorithms on the reference video data of the ETISEO project.
The internship might lead to a PhD study.
The objective of the internship is to refine the proposed evaluation metrics, to extend the existing evaluation tool, and to test the ORION platform on the reference video data.
The internship consists of both a theoretical study on performance evaluation and successful accomplishment of testing on the reference video data.
The first step will be a bibliographic study on the subject. The second step will be extension of the existing evaluation tool. The successful accomplishment consists of the evaluation of the ORION platform using the reference video data. The last step will be the performance analysis of the algorithms proposed by all the research groups involved in the ETISEO project.
| Required Background and Skills |
Strong background in C++ programming, artificial intelligence, cognitive vision.
6 months within the ORION group of INRIA Sophia-Antipolis.
François BRÉMOND et Monique THONNAT
Projet ORION, INRIA-Sophia Antipolis
2004 route des Lucioles BP 93
06902 Sophia Antipolis Cedex
Email : Francois.Bremond@sophia.inria.fr
Tel: 04 92 38 76 59
[1] Alberto Avanzi, Francois Bremond, Christophe Tornieri and Monique Thonnat,
Design and Assessment of an Intelligent Activity Monitoring Platform, in
EURASIP Journal on Applied Signal Processing, special issue in "Advances in Intelligent Vision Systems: Methods and Applications", 2005.
[2] B. Georis, F. Brémond, M. Thonnat and B. Macq,
Use of an Evaluation and Diagnosis Method to Improve Tracking Performances . The
3rd IASTED International Conference on Visualization, Imaging and Image Proceeding VIIP 03 , September 8-10 2003, Benalmadera, Spain.
[3] Etiséo, Introduction to metrics and évaluation, 2005.